Build vs Buy Retail Media Ad Tech Stack: Decision Framework and Vendor Comparison (2026)

Last updated: June 4 2026.Reviewed by Kunal Damgude, Product Marketing Specialist.

For most retail media network (RMN) operators, the honest answer to "should we build or buy our retail media ad tech stack?" is neither a pure build nor a pure buy — it is a risk-calibrated, cost-modeled choice that usually lands on buying a platform core and building selectively on top of it. Building a full in-house stack means a multi-year engineering programme, a multi-disciplinary team, and a compliance burden you own end to end; buying a platform compresses time-to-revenue to weeks but introduces vendor-dependency risk. This guide gives you the mechanics to decide with evidence rather than opinion: a weighted scoring rubric, a six-vendor capability comparison table, a total-cost-of-ownership (TCO) model, the operational failure modes that sink homegrown builds, and the hybrid patterns that let you split the difference. For the strategic overview of what each path means and the foundational components of the stack, start with our hub guide, Ad Tech Infrastructure: Building the Engine Behind Retail Media (2026 Guide); this article takes a deeper dive into the how-to-decide mechanics — the rubric, the numbers, and the vendor specifics. If you want to skip to the buy/partner path that de-risks a build, the Osmos API Hub is designed to take an RMN live in two weeks.

Why This Decision Carries More Weight in 2026

The build-vs-buy decision is no longer just an engineering-budget question — it is a survival question for differentiation. According to the IAB's 2026 guide Building a Competitive Commerce Media Ecosystem, the average brand currently buys advertising from only about six commerce media networks (CMNs). That concentration creates what the guide frames as a sorting problem: networks that fail to demonstrate clear differentiation and credible measurement risk being "implicitly sorted out of the market within the next 24–36 months." In other words, the question is not only can you stand up a retail media network but can you build one that a brand keeps in its shortlist of six.

The IAB guide organizes the competitive imperative around five strategic priorities every commerce media network should optimize for: Revenue & Profitability, Customer Experience, Market Differentiation, Measurement & Analytics (proving incremental impact by connecting media exposure to commerce outcomes), and Standardization (resolving inconsistencies in definitions and measurement). Those five priorities map directly onto the build-vs-buy calculus, because every one of them is something you must either engineer in-house or inherit from a platform. A "retail media operating system" is the full-stack software layer — ad server, supply-side and demand-side connectivity, targeting, billing, and measurement — that lets a retailer run advertising as a monetized media business rather than a side project.

The market backdrop sharpens the stakes. Retail media is "growing nearly four times faster than the total digital advertising market," according to Adtelligent's August 2025 platform launch announcement, and the category is highly concentrated at the top — Amazon takes 75.7% of current eCommerce ad spend, with Walmart second at 4.9%, per data cited in Kevel's 2026 build-vs-buy report (originally attributed to The Trade Desk). The strategic read for everyone outside the top two: you are not building to out-scale Amazon. You are building or buying to monetize your first-party traffic profitably, defensibly, and fast enough to earn a place in the brand's shortlist before the differentiation window closes. The future outlook for buying retail media ad tech, rather than building it, is shaped by exactly this pressure — time-to-differentiation now matters more than owning every line of code.

This spoke deliberately does not re-litigate the strategic overview or the high-level pros and cons of each path. For the broad, decision-level framing — the key considerations most operators weigh first — see our companion article, Build or Buy Retail Media Ad Technology: Key Considerations. From here, we go into the mechanics.

The Five Deployment Models (Not Just Two)

Treating this as a binary — build or buy — is the first mistake. In practice there are five deployment models on a spectrum, and most successful RMNs land in the middle three rather than at either extreme.

1. Pure build. A fully in-house DSP, SSP, ad server, and measurement layer, written and operated by your own engineering organization. This is the path Amazon and Walmart took, and it is the most expensive and slowest — a multi-year engineering programme even for companies whose core business is software infrastructure.

2. Managed build. An in-house team assembling and operating the stack on top of open-source and standards-based components (Prebid, OpenRTB-compatible exchanges, open ad servers). You still own the integration, scaling, and compliance work, but you are not writing an auction engine from first principles.

3. Hybrid API-first. You license a platform core and build custom layers on top of it through open APIs — typically a proprietary targeting layer, a bespoke self-serve UI, or custom reporting. This is the fastest-growing model because it preserves differentiation where it matters (your first-party data and advertiser experience) while outsourcing the commodity infrastructure (auction, ad serving, billing). The advantage of a custom retail media ad tech stack built this way is that you own the parts that distinguish you and rent the parts that do not.

4. Turnkey platform. A white-label, near-out-of-the-box stack with minimal integration. The Osmos Turnkey Solution, for example, is positioned to assemble a bespoke retail media stack in four weeks without ripping and replacing your existing systems. This model fits operators who need to start monetizing quickly and lack a dedicated ad tech engineering team.

5. Fully managed service. You outsource not just the technology but the operations — campaign management, yield, and ad ops — to a partner or managed-service provider. This minimizes internal headcount but maximizes dependency.

The architecture of a retail media platform — how the ad server, supply, demand, and measurement components fit together — is covered in depth in the hub guide; here the point is simply that your deployment model is a choice along a spectrum, and the right point on that spectrum depends on your GMV, your engineering capacity, and how much of your advantage is genuinely proprietary. The scoring rubric below turns that judgment into a structured decision.

The Weighted Scoring Rubric: A Decision Framework for RMN Operators

The single most useful artifact for a build-vs-buy decision is a weighted scorecard. The rubric below is constructed by Osmos and informed by the IAB's five 2026 strategic priorities; it is not a published third-party framework, and the weights are meant to be tuned to your situation rather than treated as fixed law. The companion considerations article covers the qualitative factors first; this rubric converts those factors into a number you can defend in a board meeting.

The method is simple:

1. Score the build path and the buy/hybrid path from 1 (poor fit) to 5 (excellent fit) on each criterion.
2. Assign each criterion a weight from 1 to 3 based on your RMN's stage and strategy.
3. Multiply score × weight, sum each column, and compare.

Here are the eight evaluation criteria — every one of them a factor practitioners cite when evaluating retail media ad tech solutions — with guidance on how a typical mid-market RMN (under roughly $500M GMV, small-to-no ad tech engineering team) tends to score them.

Read the rubric as a diagnostic, not a verdict. Two patterns recur:

• Weight time-to-revenue and compliance risk highest when you are early-stage or operate in a regulated vertical (pharmacy/health, financial services, alcohol). Here the buy/hybrid column almost always wins, because the cost of an 18-to-24-month delay — and the cost of architecting compliance from scratch — dwarfs the value of owning commodity infrastructure.
• Weight data control, customization, and strategic differentiation highest when you are a scaled operator with a proprietary signal advantage and an ad business that is a core P&L line. Here the build (or, more often, hybrid) column gains ground, because the stack itself becomes a moat — the precise argument made in our sibling piece on owning your retail media ad stack for competitive differentiation.

Crucially, the rubric rarely produces a pure-build winner for operators outside hyperscale. What it most often produces is a hybrid recommendation: buy the infrastructure, build the differentiation. The TCO model explains why.

Total Cost of Ownership: What the Build Path Actually Costs

The headline cost of an in-house build is rarely the problem operators underestimate. It is the carrying cost — the years of maintenance, compliance retrofitting, and opportunity cost that accumulate after launch. Here is how to model it honestly. (For the qualitative pros and cons of in-house building, the considerations article above covers that layer; this section is about the numbers.)

Engineering talent. Building a production ad server, a real-time bidding engine, and a measurement layer requires a multi-disciplinary team of senior ad tech engineers, data scientists, and ad ops specialists. Ad tech engineering — auction logic, sub-second serving at scale, identity resolution — is a specialized discipline that commands among the highest compensation bands in software, and the talent is genuinely scarce. (We deliberately avoid quoting a precise salary or headcount figure here: credible, independent per-role numbers for a full retail media build are not publicly established, and any specific range you see quoted should be treated as an estimate, not a benchmark.) A useful nuance from the talent market: hiring leaders often describe the broader retail media hiring gap as a training problem for campaign and ad ops roles — a media buyer can be trained into a junior retail media buyer in a few months. But that does not apply to the systems engineers who build a DSP/SSP/attribution stack; those roles are not trainable from a media-buying background, which is why the build-side talent constraint is more acute than headcount tables suggest.

Privacy and legal counsel. A build path means you own all compliance infrastructure, and that requires in-house (or dedicated outside) privacy counsel with ad tech specialization. The skills required are specific: structuring privacy-by-design into new data-processing activities, drafting service-provider agreements that flow down data-use prohibitions and opt-out obligations, and standing up the documentation that regulators now expect. This is a permanent operating cost, not a one-time legal review.

Ongoing maintenance and technical debt. The build cost is the down payment; maintenance is the mortgage. General IT industry literature commonly estimates that a large share of IT budgets — often cited around 40% — is consumed by technical-debt maintenance. (That figure is general enterprise-IT context, not a retail-media-specific benchmark, so treat it as directional.) For ad tech specifically, the debt compounds faster than typical software because the ground keeps moving: privacy regulations change, identity signals deprecate, auction logic needs tuning, and AI-driven optimization requires constant updates. Build vs buy on technical debt is really a question of who absorbs the constant change — you, or a platform amortizing it across many customers.

Migration and switching costs. If an in-house build stalls or fails to reach parity, the cost of migrating advertisers, campaigns, and historical data onto a new platform is significant — and it lands precisely when the programme is already behind.

Opportunity cost. This is the largest and most overlooked line item. Amazon's own arc shows how long the clock can run: it launched the first retail media network, the Amazon Ads Platform, in 2012, and its ad revenue grew from about $600 million in that first year to $45 billion by 2023 (Nielsen). Every quarter spent building is a quarter of ad revenue not booked and a quarter closer to the 24-to-36-month differentiation window the IAB describes.

Buy-path cost, for contrast. The buy path is not free. Platform and ad-tech fees on the buy path are typically structured as a percentage of media spend — a per-spend model that scales with revenue rather than front-loading capital. The trade-off is explicit: you convert a large, uncertain capital outlay (build) into a predictable variable cost (buy). Our own Turnkey Solution sits at the far end of that shift — on our numbers, we price it at roughly 0.003% of the cost of an in-house build.

The emerging-market wrinkle (India). Talent cost differentials can change the build calculus in markets like India. India tech compensation "runs 60 to 75% below equivalent US roles even at competitive local rates," per the India Tech Salary Report 2026 — a senior backend engineer at a global capability center in Hyderabad earns roughly ₹35–50 lakh versus ₹18–24 lakh at a mid-size IT services firm. That lower engineering cost makes a managed build more viable for large Indian e-commerce players than the US cost structure would suggest. But the savings are partly offset by compliance complexity: India's Digital Personal Data Protection Act (DPDPA), 2023, carries penalties of up to ₹250 Crore (roughly $30M USD) and obligations including consent managers, data protection officers, and 72-hour breach reporting. An Indian retailer building in-house in 2026 must architect DPDPA compliance from day one — which is exactly the kind of burden a pre-compliant bought platform absorbs.

Vendor Capability Comparison Table

If you are evaluating the buy or hybrid path, the 2026 vendor landscape divides into full-stack retail media operating systems, demand-side platforms (DSPs), supply-side/SSP-heritage platforms, and API-first infrastructure. The table below compares six options across the dimensions that matter for an RMN operator. Osmos appears first as the reference point because this is our read of the 2026 landscape as the team that builds a full-stack retail media OS — each competitor's genuine strengths are noted honestly, but assessed against what an operator actually needs to ship a network and keep it compliant, not as a neutral feature checklist.

One correction the market gets wrong constantly, and that this table fixes: Commerce Max is Criteo's demand-side platform, not Amazon's. Amazon's platform is Amazon DSP, listed as a separate column. Criteo's Commerce Max connects advertisers to retail media supply; Amazon DSP lets brands buy Amazon's inventory and audiences. They are different tools from different companies.

A few head-to-head comparisons buyers ask about most:

Criteo vs The Trade Desk. This is the most common head-to-head, and the answer hinges on platform type. Criteo's Commerce Media Platform is full-stack: its demand side (Commerce Max, live on 225+ retailers globally) plus its supply side (Commerce Yield for retailer monetization, Commerce Grid as an SSP) means a retailer can run a network on Criteo end to end. The Trade Desk is a pure-play DSP — powerful on the demand side, with strong retail-data-led measurement (Vileda lifted sales 15.6%; Magnum drove 30% incremental sales by blending weather and retail data) and UID2 for cookieless identity — but it does not provide the supply-side ad server a retailer needs. Choosing The Trade Desk as your demand partner still leaves you to build or buy your own supply-side stack. For Criteo's Commerce Max DSP features specifically: a first-price auction with CPC for sponsored products and CPM for display/video, near-real-time transaction logs, new-to-brand attribution, and full-funnel reporting across 225+ retailers.

Kevel vs Criteo, and Adtelligent vs the field. Kevel is the closest architectural cousin to Osmos's API-first approach — API-native, SaaS-priced with no revenue share, and oriented toward retailers who want to build custom experience layers on a managed core; its 2026 Adobe Experience Platform integration strengthens that position. Where it differs from a full-stack OS: it asks for more engineering investment from non-technical buyers and does not publicly disclose the compliance certifications a regulated operator needs. Adtelligent, by contrast, comes from SSP/header-bidding heritage (50,000+ publisher relationships) and launched a dedicated retail media platform in August 2025 that bundles a CDP, ad server, DOOH tooling, and attribution. It is strongest for publishers and retailers monetizing existing inventory; as a newer retail media entrant it has fewer published case studies and no disclosed pricing, onboarding timeline, or compliance certifications. As Adtelligent's VP of Retail Media, Anastasiya Shmal, put it at launch: "As retailers evolve into media companies, they need the right technology to manage and monetize their audiences effectively. Our Retail Media Ad Server is purpose-built to give them the autonomy, flexibility, and transparency retailers need."

The balanced read on the buy path. Buying is not risk-free, and Criteo's 2025 results make the point. Its retail media segment fell 17% year over year to $76M in gross revenue in Q4 2025, with $25M in spend disappearing as two clients — Uber Eats and Roundel (Target's retail media business) — reduced scope; Criteo guided full-year 2026 growth at "flat to 2%" Contribution ex-TAC. That is concrete evidence of platform-dependency and client-concentration risk on the buy side. It is also the reason the comparison should weigh vendor stability, contract terms, and exit/portability as seriously as feature checklists.

Where we differentiate in this set is breadth and disclosure rather than raw speed: full omnichannel — onsite, offsite, and in-store — in one platform (most competitors require separate layers), eight open APIs with white-label support and no rip-and-replace, and explicitly disclosed compliance (SOC 2 Type 2 and ISO 27001 certified, plus CCPA and DPDPA, and GDPR-compliant). On speed, Kevel's 14-day claim is comparable to our API Hub's two-week path — so the honest differentiator is not "fastest," it is "full-stack and compliance-disclosed."

Implementation and Operational Challenges of the In-House Build

Operators who choose to build almost always underestimate the operational, not the engineering, difficulty. Implementing a retail media platform built in-house surfaces a predictable set of failure modes — the operational challenges of a homegrown retail media solution that rarely appear in the original project plan.

Phase 1 failure mode — underestimating the supply side. A real-time auction and ad-serving system that responds in milliseconds at scale is genuinely hard infrastructure. The retail-specific demands — catalog-aware targeting across millions of SKUs, brand-safety controls, sub-second serving under load — differ materially from publisher ad serving, which is why publisher ad tech repurposed for retail tends to fall short. (Our pieces on smart ad server requirements for retail media and why retail media ad serving differs from publisher ad tech go deeper on this point.)

Phase 2 failure mode — no attribution on day one. This is the most common reason early in-house RMNs lose advertisers: without credible proof of ROAS and incrementality, brands churn. Building measurement is harder than building the ad server (the next section explains why), and shipping a network without it is shipping a product advertisers cannot justify renewing.

Phase 3 failure mode — no self-serve tooling. In-house builds frequently stall on two operational gaps: difficulty attracting mid-market advertisers without the familiar self-serve campaign tools they expect, and internal teams that cannot match the innovation velocity of dedicated platforms. Migrating advertisers from manual insertion orders to self-serve is a product and change-management problem, not just an engineering one.

The team you actually have to hire. A homegrown build requires roles most retailers do not have on staff: ad ops, yield management, privacy counsel, and brand-safety specialists, on top of the systems engineers. Each is a non-trivial, competitive hire.

Compliance buildout — and why it is structurally expensive. Building means owning the regulatory machine. Under CCPA/CPRA, businesses must conduct risk assessments for new personal-data processing activities (in force as of January 1, 2026), and automated decision-making technology (ADMT) rules — which sweep in AI-based ad targeting — become enforceable January 1, 2027. The in-house legal ad tech challenge is real: privacy-by-design has to be architected into the system, not patched on later, and that requires privacy counsel embedded in the build from the start. Retrofitting compliance into a system that was not designed for it is one of the most expensive forms of technical debt in ad tech, which is why the data-privacy regulation surface is a primary input to the build-vs-buy decision rather than an afterthought.

The plateau. In our own platform data, most RMNs stall at around 0.5% GMV monetization, while our average customer runs roughly 3x that. Whatever the exact figure, the pattern it describes — homegrown programmes plateauing well below their potential — is consistent with the failure modes above: without attribution, self-serve tooling, and continuous optimization, monetization stalls.

Attribution and Measurement: The Most Under-Estimated Build Complexity

If there is one capability that breaks in-house builds, it is measurement — and the data is stark. According to Skai's 2026 State of Retail Media Measurement and Incrementality report, only 15% of marketers report strong measurement confidence (very or extremely effective), half of brands measure incrementality at only a basic level, and just 20% are good at both measuring incrementality and applying those insights to decisions. Incrementality dominates the list of measurement challenges at 75%. Most telling for the build-vs-buy question: limited internal analytics or data science resources tops the list of barriers at 56%.

That 56% figure is the crux of why attribution is harder to build than the ad server itself. An ad server is a solved engineering problem with reference architectures; a credible measurement practice requires data science talent more than half of brands say they lack. Retail media attribution modeling and measurement is not one capability but three distinct ones — last-click attribution, multi-touch attribution, and true incrementality testing — and each demands different data infrastructure, clean-room access, and statistical expertise. Build it badly and you ship the false confidence that erodes advertiser trust; build it well and you have effectively stood up an analytics organization alongside your ad platform.

The expert consensus is that ownership, not access, is the gating factor. As Megan Conahan of Direct Agents puts it: "Lack of analytics access is the excuse; lack of ownership is the problem." And the bar advertisers now hold networks to has risen past simple ROAS. In the words of Jason Wescott of WPP Media: "The overreliance on ROAS as the benchmark of value is over. Independent, transparent measurement is the baseline." For an RMN operator, that is a direct instruction: measurement is the price of admission, and it is the part of the stack least forgiving of a from-scratch build.

Retail media attribution best practices that apply regardless of build-or-buy: insist on incrementality measurement rather than relying on last-click ROAS; ensure measurement is independent and transparent enough to satisfy sophisticated advertisers; and account for data residency and clean-room requirements so that measurement itself does not create a compliance exposure. On the build vs buy decision specifically, measurement tilts the rubric toward buy or hybrid for most operators precisely because the talent barrier (the 56%) is so high — a pre-built, audited measurement framework removes the single most failure-prone component from the critical path. For a full treatment of attribution and incrementality, our Pillar 6 measurement coverage goes deeper than this section can.

Hybrid Build+Buy Patterns: When Neither Extreme Makes Sense

For the majority of operators, the rubric points to a hybrid: buy the commodity infrastructure, build the differentiation. Three patterns recur in practice.

Pattern 1 — Buy the platform core, build the targeting layer. You license a full ad server, auction, and billing stack, and build a proprietary targeting layer on top that activates your unique first-party signal. This preserves the one thing that genuinely differentiates you (your data and how you use it) while outsourcing everything commodity. API-native platforms like Kevel and the Osmos API Hub — with eight open APIs covering campaigns, reporting, ad serving, events, advertisers, catalog, billing, and audiences — are built for exactly this.

Pattern 2 — Buy the ad server and auction engine, build the self-serve UI and reporting. Here you keep advertiser-facing experience in-house (because that is where your brand and differentiation live) and rent the serving and auction machinery beneath it. The Osmos Custom Solution is designed for this kind of partial gap-filling — enabling self-serve on top of an ad server you already love, or running multiple ad servers and multiple demand sources at once when one is not enough.

Pattern 3 — Start turnkey, graduate to API-first. A common maturity arc: launch on a turnkey platform to start monetizing in weeks, then progressively replace or extend modules via APIs as the programme matures and your engineering capacity grows. This avoids the all-or-nothing trap and lets the architecture grow with the business.

The principle underneath all three is modular composability — what is often called a "Lego-block" architecture. Rather than a monolithic rip-and-replace, you plug a platform in where you need it and keep what already works. This is also where the lock-in question lives: a genuinely API-first, modular architecture is the structural defense against vendor lock-in, because portability is designed in rather than bolted on. We cover that defense in depth in our sibling guide, how to build a scalable ad tech stack without vendor lock-in; the relevant point for hybrid patterns is that composability and lock-in avoidance are two sides of the same architectural decision.

Decision Playbook: Build, Buy, or Hybrid by RMN Profile

The rubric, TCO model, and vendor table converge on different recommendations depending on where an RMN sits. Here is the playbook by profile — covering both the advantages of buying a retail media platform and the advantages of building one, mapped to the operators each actually fits.

Profile A — Early-stage RMN (under ~$50M GMV, no ad tech team): Buy turnkey. The advantages of buying are decisive here: speed to revenue, inherited compliance, and no need to hire a specialized engineering org. A turnkey or managed deployment lets you enter the brand's six-CMN shortlist before the differentiation window closes. Building at this stage means spending your scarcest resource — time — on commodity infrastructure.

Profile B — Growth-stage RMN ($50M–$500M GMV, small engineering team): Buy API-first, build selectively. The hybrid sweet spot. License the infrastructure, then build the one or two layers (targeting, advertiser UX) where you can genuinely differentiate. This captures the advantages of building — control over your distinctive capabilities — without absorbing the cost and risk of building the whole stack.

Profile C — Scaled RMN (over ~$500M GMV, proprietary data advantage, existing engineering org): Hybrid with an owned differentiation layer. Here the advantages of building a retail media DSP or owned components start to outweigh the costs, because the stack itself can become a competitive moat and the engineering capacity exists to maintain it. Even so, most scaled operators still buy the commodity core and concentrate their build on the proprietary layer. The strategic case for owning more of the stack at this scale is made in our companion piece on owning your retail media ad stack for competitive differentiation.

Profile D — Hyperscale ($10B+ GMV, e.g., Amazon/Walmart class): Custom build can be justified — expect a multi-year programme. At this scale, a full custom build is defensible, but the timeline is measured in years, not quarters — Amazon's arc from its 2012 ad-platform launch to a $45 billion ad business is the realistic benchmark. Almost no operator outside this tier should attempt a pure build.

Emerging-market note (India and similar). Mid-tier players in markets with lower engineering costs face a genuinely different calculation: the India talent differential (60–75% below US rates) makes a managed build more viable than the US cost structure implies. But DPDPA compliance complexity — consent managers, DPOs, 72-hour breach reporting, penalties up to ₹250 Crore — pushes many mid-tier Indian operators back toward the buy/hybrid path, because a pre-compliant platform removes a from-scratch regulatory build that would otherwise have to be architected from day one.

How Osmos De-Risks the Decision

Across every profile except hyperscale, the evidence points the same direction: buy the infrastructure, build only your differentiation. Osmos is built to be that buy/partner path that de-risks a build rather than replacing the ambition behind it. The Osmos API Hub takes an RMN live in two weeks on a modular, composable, API-first architecture — eight open APIs, full omnichannel coverage (onsite, offsite, in-store), and compliance disclosed up front (SOC 2 Type 2, ISO 27001, CCPA, DPDPA, and GDPR-compliant) rather than retrofitted. Because the architecture is "Lego-block" and requires no rip-and-replace, you can buy the commodity core and build your proprietary targeting and experience layers on top — the hybrid pattern the rubric recommends for most operators. For the full platform — Adscape, ControlHub, and StratEdge unified — see Osmosphere.

The build-vs-buy decision is not a referendum on engineering pride. It is a calculation about time, talent, compliance, and differentiation — and for most RMN operators in 2026, the math favors buying the engine and building the parts that make you distinct.

Ready to skip the 18-month build?Explore the Osmos API Hub — live in two weeks.

Frequently Asked Questions

Should we build or buy our retail media ad tech stack? For most RMN operators, the answer is hybrid: buy the commodity infrastructure (ad server, auction engine, billing) and build only the layers that genuinely differentiate you (proprietary targeting, advertiser experience). A pure in-house build is realistically justified only at hyperscale ($10B+ GMV) with an existing ad tech engineering org, given multi-year timelines — Amazon's ad business, for instance, grew over more than a decade from its 2012 launch into a $45 billion engine. Use the weighted scoring rubric above — score build vs buy/hybrid across time-to-revenue, TCO, data control, compliance risk, engineering capacity, customization, scalability, and strategic differentiation — to make the call with evidence rather than instinct.

Is Commerce Max an Amazon product? No. Commerce Max is Criteo's demand-side platform (DSP), connecting advertisers to 225+ retailers globally. Amazon's platform is called Amazon DSP, which is a brand-side tool for buying Amazon's inventory and audiences — not a retail media operating system for independent retailers to build their own networks. The two are frequently confused but come from different companies and serve different purposes.

Why is attribution harder to build than the ad server? Because measurement requires data science talent that most brands lack. Skai's 2026 report found 56% of marketers cite limited internal analytics or data science resources as their top measurement barrier, and only 15% report strong measurement confidence. An ad server is a solved engineering problem; credible incrementality measurement requires statistical expertise, clean-room infrastructure, and ongoing tuning. Shipping a network without trustworthy attribution is the most common reason early in-house RMNs lose advertisers, since brands churn without proof of incremental ROAS.

What does GDPR, CCPA, and CPRA compliance mean for the build vs buy decision? Building in-house means you own all compliance infrastructure: under CCPA/CPRA you must run risk assessments for new data-processing activities (in force January 1, 2026), and automated decision-making technology (ADMT) rules covering AI-based ad targeting become enforceable January 1, 2027; GDPR adds prior opt-in consent requirements for marketing in the EU. Privacy-by-design has to be architected in, not patched on, which makes retrofitting compliance one of the most expensive forms of ad tech technical debt. A pre-compliant bought platform — one with disclosed certifications such as SOC 2 Type 2 and ISO 27001 — removes that build burden. For the privacy-first architecture overview, see the hub guide linked at the top of this article.

How much does a retail media platform cost on the buy path? Buy-path pricing is typically structured as a percentage of media spend — a variable cost that scales with ad revenue rather than a large upfront capital outlay. This contrasts with the build path's front-loaded engineering, talent, and compliance costs plus ongoing maintenance. On our own numbers, the Osmos Turnkey Solution runs at a small fraction of an in-house build — about 0.003% of the in-house build cost.

Sources

1. Kevel — Retail Media Build vs Buy: Strategy and Best Practices (updated April 30, 2026). https://www.kevel.com/blog/retail-media-build-vs-buy
2. Criteo — Commerce Max DSP. https://www.criteo.com/platform/commerce-max/
3. Criteo — Commerce Media Platform. https://www.criteo.com/platform/commerce-media-platform/
4. WMedia Research — Criteo Q4 2025 Earnings: Retail Media Troubles Overshadow Agentic Push (February 12, 2026). https://wmediaresearch.com/2026/02/12/criteo-q4-2025-earnings-retail-media-troubles-overshadow-agentic-push/
5. The Trade Desk — Retail Media. https://www.thetradedesk.com/our-demand-side-platform/retail-media
6. Adtelligent — Adtelligent Launches Retail Media Platform for Full-Funnel, Cross-Channel Retail Advertising (August 5, 2025). https://adtelligent.com/press/adtelligent-launches-retail-media-platform/
7. Skai — The 2026 State of Retail Media Measurement and Incrementality. https://skai.io/blog/the-2026-state-of-retail-media-measurement-and-incrementality/
8. Nielsen — Need to Know: What are retail media networks, and why is everybody talking about them? (2024). https://www.nielsen.com/insights/2024/what-are-retail-media-networks/
9. Pandectes — CCPA in 2026: New Requirements and Compliance Impacts You Need to Know (January 26, 2026). https://pandectes.io/blog/ccpa-in-2026-new-requirements-and-compliance-impacts/
10. SecurePrivacy — India Digital Personal Data Protection Act (DPDPA) 2023 Explained. https://secureprivacy.ai/blog/india-digital-personal-data-protection-act-dpdpa-2023
11. Kaam — India Tech Salary Report 2026 (April 28, 2026). https://www.kaam.work/blog/india-tech-salary-report
12. IAB — Building a Competitive Commerce Media Ecosystem (2026), as reported by the Path to Purchase Institute. https://p2pi.com/iab-releases-commerce-media-guidelines-focused-performance-standardization
13. Amazon Advertising — Amazon DSP. https://advertising.amazon.com/solutions/products/amazon-dsp